A data-driven intelligent partitioning method for ore blending units in open-pit polymetallic mines

By using a data-driven approach, utilizing multi-source borehole data and mathematical optimization models, and combining clustering and evolutionary algorithms, continuous polygonal boundaries are generated. This solves the problems of insufficient human experience and boundary overlap in the division of ore blending units in open-pit polymetallic mines, and achieves more efficient ore quality control.

CN123069474APending Publication Date: 2026-11-10XI'AN UNIVERSITY OF ARCHITECTURE AND TECHNOLOGY +2
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611259097.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-19
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

The current division of ore blending units in open-pit polymetallic mines relies on manual experience, making it difficult to integrate multi-source quality data and spatial engineering constraints. This results in a difficulty in balancing quality uniformity and engineering operability, and also makes it difficult to form continuous, non-overlapping operational boundaries.

Method used

A data-driven approach was adopted, which involves acquiring multi-source borehole data, identifying outliers and standardizing features, constructing a mathematical optimization model, and combining clustering and evolutionary algorithms to generate continuous polygonal boundaries, ensuring lithological consistency, spatial connectivity and area rationality.

Benefits of technology

It improves the objectivity and feasibility of dividing ore blending units, reduces quality fluctuations, ensures the continuity and non-overlap of boundaries, and is suitable for mine production scheduling.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123069474A_ABST
    Figure CN123069474A_ABST
Patent Text Reader

Abstract

This invention discloses a data-driven intelligent partitioning method for blending units in open-pit polymetallic mines. The method acquires borehole analysis and boundary data from the blast pile to be mined, completing outlier identification, spatial interpolation of missing values, feature standardization, and joint feature vector construction. It then performs preliminary partitioning of blending units based on lithology, constructs a mathematical optimization model for blending unit partitioning, builds an initial partitioning solution based on a clustering algorithm, and iteratively optimizes the partitioning based on an evolutionary algorithm. The fitness value is composed of a quality uniformity objective function and penalties for insufficient area, overlapping convex hulls, spatial disconnection, and lithological inconsistency. Optimized individuals are generated through binary tournament selection, segmented uniform crossover, and constraint-aware adaptive mutation. Finally, the borehole assignment results are reconstructed into blending unit-level polygons with complete coverage, continuous boundaries, and no overlap. This method improves the rigor, quality uniformity, and engineering feasibility of the model for partitioning blending units in open-pit polymetallic mines.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the technical fields of intelligent open-pit mining, ore quality control, spatial data processing and combination optimization, and specifically relates to a data-driven intelligent partitioning method for ore blending units in open-pit polymetallic mines. Background Technology

[0002] Open-pit polymetallic deposits are typically characterized by complex composition, associated polymetallic deposits, significant grade fluctuations, and substantial spatial heterogeneity. The blast piles awaiting mining in open-pit mines are the fundamental material source for short-term mining and ore blending plans. Treating a single blast pile as a uniform, single material unit would mask the spatial differences in lithology, metal grades, and oxidation levels within the pile, leading to fluctuations in ore quality, reduced adaptability to beneficiation processes, and decreased resource utilization efficiency. Current open-pit mine ore blending unit delineation relies heavily on manual delineation by on-site personnel based on borehole analysis data, experience boundaries, and production habits. This method suffers from high subjectivity, low efficiency, inconsistent delineation standards, and difficulty in simultaneously ensuring quality uniformity and engineering feasibility. Especially in polymetallic mines such as molybdenum, tungsten, and copper, complex coupling relationships exist between different metal grades, lithological types, oxidation rates, and harmful impurity contents. Delineation based solely on a single indicator or manual experience easily results in ore blending units with significant internal quality variations, failing to provide a reliable data foundation for precise ore blending. Some technologies employ clustering algorithms to divide boreholes or sampling points, but ordinary clustering typically only focuses on feature distances, making it difficult to guarantee that the same ore blending unit is spatially continuous, geometrically regular, and meets minimum mining requirements in terms of area. It also struggles to ensure that the unit boundary completely covers the blasting area. Therefore, there is an urgent need for a method that can integrate multi-source data such as borehole spatial location, lithology, polymetallic grade, and oxidation rate, and automatically generate ore blending units under engineering constraints such as consistent lithology, spatial connectivity, regular shape, reasonable area, and non-overlapping boundaries. This would improve the objectivity, accuracy, and feasibility of ore blending unit division in open-pit polymetallic mines. Summary of the Invention

[0003] To overcome the shortcomings of the prior art, the present invention aims to provide a data-driven intelligent division method for open-pit polymetallic mine blending units. This method overcomes the deficiencies of existing open-pit polymetallic mine blending unit division methods, which rely on manual experience, are difficult to couple with multi-source quality data and spatial engineering constraints, and are difficult to form continuous and non-overlapping work boundaries. The present invention can minimize the relative dispersion of various ore quality indicators within the same blending unit while ensuring lithological consistency. At the same time, it ensures that each blending unit is spatially connected, has an area not less than the preset minimum work area, and that convex hull boundaries do not overlap. Furthermore, it transforms discrete borehole assignment results into continuous polygonal boundaries that can be used for mine production scheduling.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A data-driven intelligent partitioning method for ore blending units in open-pit polymetallic mines includes the following steps: Step 1, Data Acquisition: Acquire multi-source borehole data and blast pile boundary data of the open-pit polymetallic mine to be mined; the multi-source borehole data includes borehole number, borehole plane coordinates, bench height, lithology type and more than one ore quality index, the ore quality index includes one or more of the following: metal grade, oxidation rate and harmful impurity content.

[0005] Step 2, data preprocessing to obtain joint feature vector: outlier identification, missing value spatial interpolation and feature standardization are performed on the multi-source borehole data to obtain standardized spatial features, standardized quality features and a joint feature vector formed by weighted fusion of the two.

[0006] Step 3, construct a mathematical optimization model for dividing ore blending units: establish a mathematical optimization model with borehole number and number of ore blending units as decision variables. The model takes minimizing the variation in ore quality within the ore blending unit as the objective function and uses the geometric shape, planar size, spatial connectivity, and non-overlapping boundaries of the ore blending unit as constraints.

[0007] Step 4, preliminary division of hierarchical ore blending units based on lithology: The borehole samples are divided into primary zones according to lithology type. Borehole samples with the same lithology type are divided into the same lithology zone. For each lithology zone, the area of ​​the convex hull formed by the borehole points in the lithology zone is calculated. If the convex hull area is less than the preset minimum convex hull area threshold, the lithology zone is regarded as an ore blending unit. Otherwise, proceed to the next step and further subdivide the lithology zone.

[0008] Step 5, Initial solution construction based on clustering algorithm: Based on the joint feature vector, an initial ore blending unit division scheme is constructed in each lithological partition that needs to be subdivided through a clustering algorithm; the number of clusters is adaptively determined according to the total area of ​​the convex hull constructed by all boreholes in the blast pile, the minimum area threshold of the convex hull constructed by boreholes in a single ore blending unit, and the effective clustering ratio; and the spatially discrete clustering results are split, recoded, and the discrete point assignment is adjusted using DBSCAN connected component detection.

[0009] Step 6: Iterative optimization of ore blending unit division based on evolutionary algorithm: Fixed-length integer encoding is used to represent the affiliation relationship between boreholes and ore blending units. Decoding constraints are corrected for the encoded individuals. Fitness values ​​containing quality uniformity objective function and constraint penalty terms are calculated. Selection, crossover and mutation operations are performed to obtain the optimized borehole affiliation number.

[0010] Step 7, Ore Blending Boundary Reconstruction: The borehole assignment results from Step 6 are transformed into continuous polygonal boundaries for ore blending boundary reconstruction. The boundary reconstruction module constructs a Voronoi diagram using all borehole points as generators, finiteifies the unbounded Voronoi elements, trims them with the blast pile boundary polygons, and calculates the union of borehole-level Voronoi elements according to the ore blending unit to which the borehole belongs, thus obtaining the ore blending unit-level polygonal boundary.

[0011] Step 8, Output Results: Output the ore blending unit number, borehole attribution, ore blending unit boundary coordinates, area, lithology type, quality statistics, and evaluation indicators; the ore blending unit boundary coordinates can be used in GIS systems, mine scheduling systems, or on-site positioning systems.

[0012] Compared with the prior art, the beneficial effects of the present invention are: First, this invention transforms the division of open-pit polymetallic ore blending units from manual experience-based delineation into a quantitative division process driven by multi-source borehole data. This process can simultaneously utilize information such as borehole spatial coordinates, lithological type, polymetallic grade, oxidation rate, and harmful impurity content, thereby improving the objectivity and repeatability of ore blending unit division.

[0013] Second, this invention uses lithological consistency as a primary hard constraint and performs subdivision optimization within a single lithological zone to avoid fluctuations in the beneficiation process caused by cross-lithological mixing and improve the consistency of geological properties of the ore blending unit.

[0014] Third, this invention divides the ore blending unit into a combinatorial optimization problem with constraints such as spatial connectivity, convex hull area, non-overlapping convex hull boundaries, and single borehole affiliation. By solving the initial solution through clustering and evolutionary optimization in a coordinated manner, the quality fluctuation within the unit is reduced while ensuring the operability of the project.

[0015] Fourth, this invention reduces spatial discrepancies, insufficient area, or boundary conflicts caused by clustering and evolutionary operations through DBSCAN connected component detection, minimum area correction, and convex hull overlap correction, making the output unit more suitable for loading, transportation, and on-site boundary management.

[0016] Fifth, this invention utilizes Voronoi diagrams to reconstruct the boundaries of borehole-level spatial attribution, which can convert discrete borehole labels into ore-mixing unit-level polygons with complete coverage, continuous boundaries, and no spatial overlap, making it easy to directly import into GIS systems, mine intelligent scheduling systems, or on-site positioning systems.

[0017] Sixth, the ore blending unit output by this invention includes information such as boundary coordinates, area, average quality and quality dispersion, which can be used for short-term ore blending planning, ore quality prediction, equipment operation area delineation and electronic fence generation, thereby improving the execution accuracy of precise ore blending in open-pit polymetallic mines from the source. Attached Figure Description

[0018] Figure 1 This is a flowchart of the present invention.

[0019] Figure 2 This is a schematic diagram of the ore blending unit division problem in an embodiment of the present invention, illustrating the process of generating multiple independent ore blending units from discrete borehole data. Detailed Implementation

[0020] The present invention will be further described below with reference to the accompanying drawings and embodiments. The following embodiments are used to illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention. Without departing from the concept of the present invention, those skilled in the art can make equivalent substitutions or conventional adjustments to the parameter values, software implementation methods, and data interface forms.

[0021] This invention provides a data-driven intelligent partitioning method for open-pit polymetallic mine ore blending units. This method is applicable to open-pit molybdenum-tungsten-copper mines, tungsten-molybdenum mines, copper-molybdenum mines, rare metal mines, and other open-pit polymetallic mines with multiple quality indicator control requirements. Input data can come from borehole databases, sampling and testing systems, mine 3D geological databases, or production scheduling systems.

[0022] The method described in this invention is executed collaboratively by a data acquisition module, a data preprocessing module, a mathematical optimization model construction module, a preliminary lithological level division module, an initial solution construction module, an evolutionary optimization module, a boundary reconstruction module, and a result output module. Figure 1 and Figure 2 As shown, the specific steps are described below: Step 1, Data Acquisition.

[0023] Read the boundary polygon of the blast pile to be mined and the set of borehole samples. Each borehole sample includes at least the borehole number, east coordinate, north coordinate, bench height, and lithology type; preferably, each borehole sample also includes Mo, WO3, Cu, Fe, oxidation rate, sulfur content, or other quality indicators related to the mineral processing process. The blast pile boundary polygon is used to define the subsequent Voronoi boundary trimming range.

[0024] Step 2, data preprocessing.

[0025] Step 2.1, outlier removal.

[0026] A statistical screening method based on the interquartile range (IQR) was used to detect outliers in the grade analysis values ​​of various components of the ore. Specifically, for a grade sequence of a certain metal component, its lower quartile was calculated. and the upper quartile It will fall Observations outside the range are marked as anomalous.

[0027] Step 2.2, Spatial interpolation of missing data. Outliers and original missing values ​​are used as imputation values, and inverse distance weighted (IDW) spatial interpolation is performed on the missing data. (This is for the metric.) j ,drilling p i The formula for calculating the missing indicator value is as follows: (1) (2) (3) in, It is the first i The first missing borehole j Estimated values ​​of various quality indicators; It is the known number l The first borehole j Quality index values; N i This is the number of adjacent boreholes used for interpolation, and its value is 5. w il It is a weight based on spatial distance; d il It is the first i The first borehole and the first l The Euclidean distance between the boreholes; and They are the first i The first borehole and the first l The geodetic coordinates of each borehole are denoted as follows: and .

[0028] Step 2.3, feature standardization.

[0029] z-score standardization was applied to the borehole plane coordinates and ore quality indicators respectively to obtain standardized spatial features and standardized quality features. Z i =[ w s · Pos i , w q · Q i Construct the first i The joint feature vector of the boreholes, where Pos i To standardize spatial features, Q i To standardize quality characteristics, w s For spatial feature weights,w q As the weight of quality characteristics, and w s + w q =1.

[0030] Step 3: Construct a mathematical optimization model for dividing the ore blending units.

[0031] Based on the macroscopic division of the blast pile to be mined according to the borehole lithology, the problem of further refining the ore blending unit division is transformed into a solvable combinatorial optimization model with multiple spatial constraints. The optimization model is as follows: (1) Index set : I represents the set of indices of the boreholes in the blast pile to be mined. Indicates the first i Index of each borehole, N This represents the total number of boreholes drilled.

[0032] J represents the amount contained in the blasting pile to be mined. M A set of indices for various qualities (such as molybdenum, copper, iron, and oxidation rate). Indicates the first j An index of quality.

[0033] Rock represents the contents contained in the unmined blast pile. R A set of indexes for various rock types (such as felsic hornfels, diopside-garnet skarn, and diopside felsic hornfels). Indicates the first r Index of rock types.

[0034] :P represents the unmined blast pile N A collection of drill holes, Indicates the first i One hole.

[0035] K represents the set of indices for all ore blending units. K It is the total number of ore blending units. Indicates the first k Index of each ore blending unit.

[0036] C represents the set of all ore blending units, where C k For the first k The set of borehole indices within a single ore blending unit satisfies and .

[0037] (2) Variables Indicates the first i Geodetic coordinates of the borehole E i East coordinates, N i The coordinates are north.

[0038] Indicates the first i One hole M Via ore quality characteristic vectors (such as Mo, WO3 content, etc.), where q ij Indicates the first i The first borehole j The quality index values.

[0039] rock i : indicates the first i The rock type of each borehole .

[0040] (3) Decision variables K : Represents the total number of ore blending units that divide a complete blast pile into.

[0041] : Indicates the first i Index of the ore blending unit to which each borehole belongs.

[0042] (4) Optimization objective: Minimize the differences in quality indicators within the unit. This is achieved by minimizing the sum of the differences in quality indicators within all blending units. The smaller the target value, the more uniform the ore quality within the blending unit, and the greater the differences in ore quality between blending units.

[0043] (4) (5) (6) in, k It is an index of the ore blending unit. It belongs to the first k The set of boreholes for each ore blending unit is labeled as follows: , K It is the total number of ore blending units. Indicates the first i Index of the ore blending unit to which each borehole belongs. Indicates the first i One hole, N This represents the total number of boreholes. It is the first k Total number of boreholes within each ore blending unit; This is an indicator function that takes the value 1 if the condition is met, and 0 otherwise. q ij Indicates the first i The first borehole j The quality index values, It is the first k In the first ore blending unit j The average value of various quality indicators, M It is the total number of quality indicators.

[0044] (5) Constraint 1: Regular geometric shape constraint of ore blending unit.

[0045] Convex hulls are used to control the geometry and spatial connectivity of ore blending units. A convex hull contains no concave regions, and its boundaries are smooth or composed of convex polygonal lines. This regular geometry is highly compatible with the operational modes of large mining equipment. According to the definition of a convex hull, for any ore blending unit, its regular geometry must satisfy the following three constraints: 1) The number of boreholes in this ore mixing unit must be at least 3 to meet the requirement of constructing a convex hull using the convex hull algorithm. This constraint is expressed as follows: (7) 2) All boreholes within this ore blending unit must be within the constructed convex hull. Within this constraint, it is expressed as follows: (8) 3) The line connecting any two drilled holes within the constructed convex hull must be completely contained within the convex hull. This constraint is expressed as follows: (9) in, p and q It is the index of any two holes drilled within the convex hull. and Drill holes separately p and q coordinates t These are the combination coefficients.

[0046] (6) Constraint 2: Planar dimension constraint of the ore blending unit.

[0047] A minimum area constraint is set for each independent ore blending unit, meaning the spatial coverage area of ​​each unit must not be less than a preset minimum engineering mining area threshold, in order to balance the requirements of unit homogenization with production practicality. The calculation formula is as follows: (10) in: Indicates according to the first k The convex hull generated by the borehole coordinates within each ore blending unit. Represents the convex hull The area refers to the spatial coverage area of ​​the ore blending unit. A min This represents the minimum area threshold of the convex hull constructed by all boreholes within a single ore blending unit, i.e., the minimum engineering mining area threshold.

[0048] The calculation steps are as follows: First, using the convex hull algorithm, based on the... k Geodetic coordinates of all boreholes within a single ore-mixing unit Constructing the convex hull To obtain the vertices of the convex polygon. ,in L Represents the convex hull The total number of vertices, Represents the convex hull The l Vertex. The coordinates are as .

[0049] Secondly, the convex hull is calculated using the polygon area formula. The area is as follows: (11) in, and These represent the coordinates of the vertices of the convex polygon. and This is because it forms a closed polygon, so the first vertex coincides with the last vertex.

[0050] The minimum area of ​​the convex hull is the lower limit of the minimum effective area allowed for the convex hull of a ore blending unit, i.e., the minimum engineering mining area threshold. Each ore blending unit corresponds to only one convex hull, and has only one convex hull area; there are no multiple convex hull areas. The purpose of setting this constraint is to avoid dividing the ore blending units into extremely small fragmented units; such excessively small units cannot be used for loading operations on site and are not feasible for engineering implementation.

[0051] (7) Constraint 3: Spatial connectivity of the ore blending unit.

[0052] A single ore mixing unit should be a spatially connected and complete area on the blasting plane; it is not permissible for the same unit to consist of multiple isolated fragments. This invention employs 8... Neighborhood connectivity criteria are used to determine the spatial connectivity of ore blending units. Let ore blending unit... Set of sample points within a unit If there exists a path between any two boreholes within the same ore blending unit, formed by sampling points within the unit, and adjacent sampling points on the path satisfy neighborhood relationships in eight directions (up / down, left / right, and diagonal), then the ore blending unit satisfies spatial connectivity constraints, expressed mathematically as follows: in, , For ore blending unit Any two sampling points within; For connection and All intermediate sampling points on the connecting path belong to the ore blending unit. ;symbol This means that two points satisfy 8 Neighborhood spatial adjacency relationship. This formula means that any two points within a cell can be connected via 8 sampling points within the cell. Neighboring paths reach each other.

[0053] (8) Constraint 4: The boundaries of the ore blending unit have no overlap constraint.

[0054] The minimum convex polygon constructed using the convex hull algorithm is used as the spatial representation of the ore blending unit. By strictly limiting the spatial relationships of the convex hull boundaries of different units, boundary intersections and regional overlaps between units are geometrically eliminated, providing a clear and conflict-free spatial basis for subsequent mining planning, ore allocation, and other processes. The convex hulls of any two ore blending units have no spatial intersection, and their mathematical formula can be expressed as follows: (12) in, and They respectively represent the inclusion and The bulges of all internal drilled holes, This indicates that the two convex hulls have neither overlapping interior regions nor shared boundary points (i.e., no spatial point belongs to both convex hulls at the same time). This indicates that the boreholes within all units together constitute a complete coverage of the boreholes within the entire blast pile to be mined.

[0055] (9) Constraint 5: Lithological consistency constraint of ore blending unit All boreholes within each ore blending unit must belong to the same lithology; that is, the same ore blending unit cannot contain boreholes of different lithology types. Therefore, the lithology consistency constraint can be expressed as: (13) In the formula, i and l For the same ore blending unit Ck The index of any two boreholes.

[0056] (10) Constraint 6: Decision variable constraints.

[0057] For the decision variable of the number of ore blending units K Its constraints are: (14) In the formula, K min This represents the minimum number of ore blending units. K max This represents the maximum number of ore blending units.

[0058] For borehole allocation decision variables x i Its constraints are: (15) In the formula, x i Indicates the first i The constraint assigns each borehole to a specific ore blending unit number. This constraint ensures that each borehole in the blast pile must be assigned to a specific ore blending unit, and that the assigned unit number must be within a valid range.

[0059] Step 4: Preliminary division of stratified ore blending units based on lithology.

[0060] First, the borehole samples are coarse-grained grouped according to lithology, and then fine-grained ore blending unit optimization is performed within each group. The specific steps are as follows: Step 4.1: Traverse the lithology of each borehole sample and classify borehole samples with the same lithology into the same lithology zone.

[0061] Step 4.2: Construct a convex hull based on the planar coordinates of the borehole samples within each lithological zone, and calculate the area of ​​the convex hull.

[0062] Step 4.3: If the area of ​​the convex hull is less than the preset minimum convex hull area threshold... A min If so, the corresponding lithological zone will be used as a single ore blending unit.

[0063] Step 4.4: If the area of ​​the convex hull is not less than the preset minimum convex hull area threshold... A min Then, the corresponding lithological zoning will be input into the initial zoning module for further subdivision, so that boreholes of different lithological types are not assigned to the same ore blending unit.

[0064] Step 5: Constructing the initial solution based on the clustering algorithm.

[0065] An initial ore blending unit partitioning scheme is constructed within each lithological subdivision. K-means clustering is performed based on joint feature vectors. The number of clusters is adaptively determined according to the total area of ​​the convex hull constructed from all boreholes within the blast pile, the minimum area threshold of the convex hull constructed from boreholes within a single ore blending unit, and the effective clustering ratio. DBSCAN connected component detection is used to split, recode, and adjust the discrete point assignments of the spatially discrete clustering results. The specific steps are as follows: Step 5.1, clustering input with multi-feature fusion.

[0066] A weighted fusion mechanism is introduced during the feature space construction stage to combine spatial coordinates and quality features into a unified feature vector according to certain weights. The weighted combined feature vector is constructed as follows: (16) in, w s and w q These are spatial feature weights and quality feature weights, respectively, satisfying... w s + w q =1.

[0067] Based on the combined feature vectors, the objective function of K-means clustering is expressed as: (17) in, C k Indicates the first k A collection of boreholes for each ore blending unit. for C k The centroid in the combined feature space. yes C k Inner i The feature vector is a weighted combination of borehole spatial coordinates and quality features.

[0068] Step 5.2: Adaptively determine the candidate interval for the number of clusters.

[0069] Adaptive K The value determination strategy is implemented through three stages: range pre-setting, constraint estimation, and validity screening. The specific steps are as follows: Step 5.2.1, Initial K Value range preset.

[0070] Based on domain experience, pre-defined K The initial search range for the value is This scope takes into account the actual needs of dividing the ore blending units. It should not be too small, which would result in overly large units, or too many, which would lead to management complexity. It also leaves room for future adjustments.

[0071] Step 5.2.2, based on area constraints K Value range correction.

[0072] The total area of ​​the convex hull of the computation region is A total Based on minimum area constraint A min The theoretical upper bound for the number of clusters can be estimated using the total area of ​​the convex hull of the regions. Therefore, the upper bound for the number of clusters is: (18) in, The upper limit of clustering, This is the area utilization coefficient, and its value is determined to be 0.8 based on actual engineering practice. The reason for introducing this coefficient is that, since there is a certain distance between the boreholes, the area of ​​the bulge formed by the boreholes inside the ore blending unit is definitely smaller than the actual area of ​​the ore blending unit.

[0073] The corrected range of K values ​​is: (19) Step 5.2.3, Cluster validity screening.

[0074] Define valid clustering: If clustering C k convex hull area ,but C k For effective clustering, let's call it . , k =1,2,… K Otherwise, it is 0. Formula for calculating the effective clustering ratio: (20) For each initial K Considering the sensitivity of the K-means algorithm to initial centroids, we performed 10 independent clustering attempts using different random seeds, retaining the largest... r ( K The corresponding clustering results.

[0075] exist Within, retain the cluster with the largest effective cluster ratio. K The value is used as the input parameter for K-means clustering.

[0076] (twenty one) Step 5.3, Clustering unit spatial connectivity correction.

[0077] Step 5.3.1: Calculate the spatial distance of borehole samples within the cluster.

[0078] Cluster C k Include n k One borehole sample, cluster C k Middle i The standardized spatial coordinates of each borehole sample are: Drilling samples i and j The Euclidean distance between them is: (twenty two) Step 5.3.2, determine the neighborhood radius of DBSCAN.

[0079] Cluster C k For each borehole sample, calculate its top 3 nearest neighbors ( k =3, the mean distance for adapting to 2D spatial connectivity determination) (twenty three) Among them, top-3NN ( ki ) = 3 , Indicates borehole sample i The set of the top 3 nearest neighbor borehole samples is adapted for 2D spatial connectivity determination.

[0080] The average nearest neighbor distance at the cluster level is: (twenty four) Considering the slight spatial dispersion of the ore mixing borehole samples, the neighborhood radius is... Set it to twice the average nearest neighbor distance at the cluster level, that is: .

[0081] Step 5.3.3: Determine the minimum number of samples for DBSCAN.

[0082] Combining the requirements of convex hull construction (requiring at least 3 borehole samples) and spatial connectivity, the minimum number of samples is set to 3.

[0083] Step 5.3.4: Perform connectivity component detection on the initial clustering units.

[0084] Clusters are analyzed using the DBSCAN algorithm. C k Perform density clustering and output the connected component labels for each borehole sample. , m The number of connected components. If Then determineC k There exist spatial discrete subclusters; if m =1, then C k It meets the requirements for spatial connectivity.

[0085] Step 5.3.5, Discrete sub-cluster splitting and cluster ID recoding.

[0086] If DBSCAN outputs multiple connected components, each connected component containing at least 3 boreholes is split into independent candidate units, and the candidate units are re-encoded. For discrete point sets with fewer than 3 boreholes, the distance to the centroid of each valid candidate unit is calculated, and the point is assigned to the nearest valid candidate unit.

[0087] Step 6: Iterative optimization of ore blending unit division based on evolutionary algorithm.

[0088] Let the set of borehole samples inside the blasting pile be represented as ,in The total number of boreholes; the characteristics of a single borehole sample are represented as follows: p i = ( E i , N i , rock i , q i1 , ..., q iM The number of ore blending units is expressed as follows: The set of ore blending unit IDs is represented as follows: Each ID corresponds to an independent mining unit. The detailed optimization steps are as follows: Step 6.1, fixed-length integer encoding based on cluster ID.

[0089] Each individual corresponds to a complete ore blending unit division scheme, using fixed-length integer coding. The code length is equal to the total number of boreholes in the blast pile to be mined, and each gene bit is taken as the ID of the ore blending unit to which the borehole belongs, ensuring a one-to-one correspondence between the code and the division scheme, while taking into account both coding simplicity and operational flexibility.

[0090] Define individual coding as 3D integer vector: (25) Among them, gene locus Indicates the first Drilling holes The ID of the ore blending unit to which it belongs must meet the following constraints: ①Single-homing constraint: That is, each borehole sample belongs to only one ore blending unit; ② No empty element constraint: That is, each ore blending unit must contain at least one borehole sample to avoid empty units; ③ Fixed length constraint: Encoding length It matches the total number of boreholes and is suitable for datasets of different sizes.

[0091] Step 6.2, constraint correction decoding.

[0092] Decoding is the process of converting the encoded vector into a single data structure. The process of mapping to an actual ore blending unit partitioning scheme involves core steps such as code parsing, constraint verification, and invalid code correction. This ensures that the decoding results meet engineering constraints such as spatial continuity, minimum area, and geometric shape. The process is as follows: Step 6.2.1: Code parsing to generate the initial ore blending unit division scheme.

[0093] Based on encoding vector The initial set of ore blending units will be obtained by grouping the units by borehole ID: , among which, the The borehole sample set for each ore blending unit is as follows: .

[0094] Step 6.2.2, Spatial Continuity Verification and Correction For each initial unit DBSCAN is used to detect connected components, and the set of connected components is obtained. , m The number of connected components.

[0095] The connected component with the most borehole samples is selected as the core sub-cluster. The remaining discrete sub-clusters Drilling samples Correcting attribution: (26) in, For unit The centroid is calculated using the following formula: (27) (28) The Euclidean distance from the borehole sample to the centroid: (29) Step 6.2.3, verification and correction of non-overlapping constraints at convex hull boundaries.

[0096] After connectivity correction, the unit set is obtained. Then, for each unit C k Pregenerated convex hull H k .

[0097] For all convex hulls { H 1, H 2,…, H K Perform pairwise overlap verification. Calculate the intersection area of ​​the pairwise convex hulls. S kl =Area( H k H l ).like Then, a new ownership assignment will be performed on the overlapping boreholes: (30) in For the unit centroid, Given the Euclidean distance from the sample to the centroid, assign the sample to a closer cell to eliminate initial overlap. Update the cells and convex hulls, and repeat the check until all convex hulls are non-overlapping.

[0098] Step 6.2.4, Minimum area constraint verification and correction.

[0099] Calculate each unit The area of ​​the convex hull is given by the formula: (31) in, for The set of vertices of the convex hull. (Closed convex hull) Let be the number of vertices of the convex hull.

[0100] like Then all borehole samples of that cell will be merged into the nearest valid cell: (32) Step 6.2.5: Generate the revised partitioning scheme.

[0101] After the above verification and correction, a ore blending unit partitioning scheme that satisfies all constraints is obtained, completing the mapping from coding to the actual partitioning scheme.

[0102] (33) Step 6.3, Calculate the fitness evaluation function The fitness evaluation function uses ore quality uniformity as a quantitative indicator, and its calculation formula is as follows: (34) The fitness function outputs a non-negative continuous value, and the smaller the value, the better the ore blending unit division scheme.

[0103] Step 6.4: Execute the binary tournament selection mechanism.

[0104] set up For the first Generational population; Popsize Population size; For the first The encoding vector of each individual, which is a fixed-length integer, corresponds to the affiliation relationship between the borehole and the ore blending unit; For individuals The smaller the fitness value, the better the ore blending unit division scheme; To select candidate offspring after the operation, the steps for implementing the binary tournament selection mechanism are as follows: Step 6.4.1: Initialize the candidate population to be empty, i.e. ; Step 6.4.2, from the current population Two distinct individuals are randomly selected from the pool to form a subset of the tournament: And they are mutually exclusive; Step 6.4.3: Select the individual with the smallest fitness value within the subset to add to the candidate population; (35) }(36) Step 6.4.4: Repeat Steps 2 and 3 until the candidate population size reaches the target size. Popsize ; Step 6.4.5: Output candidate population , as input for crossover / mutation operations.

[0105] Step 6.5: Perform the segmented uniform crossover operation.

[0106] set up , For the two parent individuals involved in the crossover; N This represents the total number of boreholes. For the spatial segmentation set of borehole samples, To determine the number of segments, based on the borehole sample Equal interval division of coordinates; ; For the first The x-coordinate of each borehole sample and The extreme values ​​of the sample's x-axis; The crossover probability controls the frequency of crossover operations. and These are the two offspring individuals generated by crossover. The steps for the segmented uniform crossover operation are as follows: Step 6.5.1: Randomly pair individuals in the candidate population, and record the pairing results as... ; Step 6.5.2, traverse each spatial segment Generate intra-segment cross mask Where 0 represents inheritance from the parent generation. The gene, 1 indicates inheritance from the father. The genes in the mask have a 0 / 1 ratio of 1:1; Step 6.5.3, for each gene locus within the segment Generate offspring genes: (37) Step 6.5.4: Output the offspring population after crossover. .

[0107] Step 6.6: Perform constraint-aware adaptive mutation operation.

[0108] set up Individuals awaiting mutation; For the first The probability of mutation in a generation; For the sample A spatial neighborhood sample set; For the sample The set of cluster IDs within the neighborhood; The set of all valid cluster IDs; The number of ore blending units; For the mutated individual.

[0109] Constraint-aware adaptive mutation operation traverses each individual in the offspring population after crossover. For each gene locus Perform the following operations: Step 6.6.1: Calculate the adaptive mutation probability for the current iteration. Calculate the adaptive mutation probability using the exponential decay formula: (38) in The initial mutation probability, This represents the maximum number of iterations. The mutation probability decays exponentially in the later stages of iteration, significantly reducing the probability of destroying high-quality convex hull structures and preserving stable unit morphologies in the later stages of optimization. Step 6.6.2, Extract Samples Cluster ID set within the neighborhood ,like Then take the set of all valid cluster IDs. ; Step 6.6.3, from Randomly select a cluster ID as the mutated value: (39) The mutation only selects the unit ID within the sample space neighborhood, and does not randomly change the encoding globally, thus avoiding the tearing of the convex hull contour and the induction of unit boundary intersection and overlap from the root cause of long-distance gene modification. Step 6.6.4, verify the number of samples corresponding to the mutated cluster ID: if the number of samples for the ID is less than 3, the convex hull cannot be constructed, and the cluster ID is reselected until the requirement is met; Step 6.6.5: Output the mutated population. As input for the next generation of the population.

[0110] Step 7: Reconstruct the boundaries of the ore blending units based on Voronoi diagram technology. The specific steps are as follows: Step 7.1: Construct a borehole-level Voronoi diagram using all borehole points as generators.

[0111] Suppose that there exists a certain blasting pile to be mined. A set of drilling points , where each point This indicates the east and north coordinates of the borehole.

[0112] For generator Its corresponding Voronoi element is defined as: (40) in The distance is Euclidean, i.e.: (41) All Voronoi units This constitutes a complete, gapless partition of the plane, where adjacent Voronoi elements share a boundary segment, i.e., the perpendicular bisector of two adjacent generators, satisfying the following property: (42) Step 7.2: Merge the Thiessen polygons at the ore blending unit level.

[0113] After the borehole partitioning clustering results output by the evolutionary algorithm, each borehole point Already allocated to the ore blending unit ( To obtain continuous boundaries at the ore blending unit level, Voronoi cells of all borehole points within the same ore blending unit are merged. Let the... The set of borehole indexes contained in each ore blending unit is as follows: Then the Thiessen polygon region corresponding to the ore blending unit The following is obtained by finding the union of the Voronoi elements of each borehole: (43) Since the boundaries of each borehole Voronoi cell are already strict nearest neighbor dividing lines, adjacent Voronoi cells within the same ore-mixing unit are precisely connected at the common boundary, and their union operation results in a connected polygon without internal holes.

[0114] Step 8, output the results.

[0115] The data includes the ore blending unit number, its lithology, borehole numbers, boundary vertex coordinate sequence, ore blending unit area, mean values ​​of each quality index, dispersion of each quality index, minimum area constraint satisfaction, spatial connectivity verification results, and boundary overlap verification results. The output format can be a tabular file, a GIS vector file, a database record, or mine scheduling system interface data.

[0116] The technical solutions of this invention are not limited to the above embodiments. For example, K-means clustering can be replaced by a constrained clustering algorithm with equivalent functionality; the DBSCAN neighborhood radius multiple can be adjusted according to the borehole mesh size; the evolutionary algorithm can also employ genetic algorithms, differential evolution, particle swarm optimization, or other heuristic algorithms capable of handling integer encoding; Voronoi boundary trimming can be performed using a GIS geometry library, spatial database, or mining 3D software. As long as the above substitutions still achieve data acquisition, data preprocessing, preliminary division of lithology-based hierarchical ore blending units, construction of a mathematical optimization model for ore blending unit division, initial solution construction based on clustering algorithms, iterative optimization of ore blending unit division based on evolutionary algorithms, and Voronoi continuous boundary reconstruction, they all fall within the scope of protection of this invention.

Claims

1. A data-driven intelligent partitioning method for open-pit polymetallic mine ore blending units, characterized in that, Includes the following steps: Step 1: Obtain multi-source borehole data and blasting pile boundary data of the open-pit polymetallic mine to be mined, and perform preprocessing to obtain a joint feature vector; the multi-source borehole data includes borehole number, borehole plane coordinates, bench height, lithology type and more than one ore quality index, the ore quality index includes one or more of metal grade, oxidation rate and harmful impurity content. Step 2: Establish a mathematical optimization model with borehole number and number of ore blending units as decision variables. The model takes minimizing the variation in ore quality within the ore blending unit as the objective function and uses the geometric shape, planar size, spatial connectivity, and non-overlapping boundary of the ore blending unit as constraints. Step 3: Divide borehole samples with the same lithological type into the same lithological zone, calculate the convex hull area formed by the borehole points in each lithological zone, and if the convex hull area is less than the preset minimum convex hull area threshold, then the lithological zone is regarded as a mineral blending unit; otherwise, proceed to the next step to further subdivide the lithological zone. Step 4: Based on the joint feature vector, construct an initial ore blending unit division scheme within each lithological partition that needs to be subdivided using a clustering algorithm; Step 5: Perform iterative optimization of ore blending unit division based on evolutionary algorithm to obtain optimized borehole assignment numbers; Step 6: Convert the borehole assignment results from Step 5 into continuous polygon boundaries, reconstruct the ore blending boundaries, and output the ore blending unit number, borehole assignment relationship, ore blending unit boundary coordinates, area, lithology type, quality statistics, and evaluation indicators.

2. The data-driven intelligent partitioning method for open-pit polymetallic mine ore blending units according to claim 1, characterized in that, The preprocessing involves identifying outliers, spatially interpolating missing values, and standardizing features in the multi-source borehole data to obtain standardized spatial features, standardized quality features, and a joint feature vector formed by weighted fusion of the two.

3. The data-driven intelligent partitioning method for open-pit polymetallic mine ore blending units according to claim 1, characterized in that, In step 2, the objective function of the model is expressed as: in, It belongs to the first k A collection of boreholes for each ore blending unit. K This is the total number of ore blending units; It is the first k The total number of boreholes within each ore blending unit Indicates the first i Index of the ore blending unit to which each borehole belongs. Indicates the first i One hole, N This represents the total number of boreholes. This is an indicator function that returns 1 if the condition is met, and 0 otherwise. ,q ij Indicates the first i The first borehole j The quality index values, It is the first k In the first ore blending unit j The average value of various quality indicators, M It is the total number of quality indicators; For any ore blending unit, its geometric constraints include: The number of boreholes in the ore blending unit is at least three, in order to meet the requirement of constructing a convex hull using the convex hull algorithm. All boreholes within the ore blending unit must be within the constructed convex hull; The line connecting any two drilled holes within the constructed convex hull must be completely contained within the convex hull. For any ore blending unit, its planar dimensional constraints are as follows: The spatial coverage area of ​​the ore blending unit shall not be less than the preset minimum engineering mining area threshold; The spatial connectivity constraint is: A single ore mixing unit should be a complete, spatially connected area on the blasting plane; it is not allowed for the same unit to be composed of multiple isolated fragments. The boundary non-overlapping constraint is: The convex hulls of any two ore blending units do not intersect in space.

4. The data-driven intelligent partitioning method for open-pit polymetallic mine ore blending units according to claim 3, characterized in that, The constraints on the objective function also include lithological consistency constraints of the ore blending unit and constraints on decision variables; The lithological consistency constraint of the ore blending unit means that all boreholes within each ore blending unit must belong to the same lithology, that is, the same ore blending unit must not contain boreholes of different lithological types. The decision variable constraints, for the decision variable of the number of ore blending units K Its constraints are: For borehole allocation decision variables x i Its constraints are: In the formula, K min This represents the minimum number of ore blending units. K max This represents the maximum number of ore blending units. x i Indicates the first i The ore blending unit number to which each borehole belongs.

5. The data-driven intelligent partitioning method for open-pit polymetallic mine ore blending units according to claim 3, characterized in that, The spatial coverage area of ​​the ore blending unit is the area of ​​the convex hull generated by the borehole coordinates within the ore blending unit. The calculation steps are as follows: First, using the convex hull algorithm, based on the... k Construct a convex hull using the geodetic coordinates of all boreholes within a ore blending unit to obtain the vertices of the convex polygon; Secondly, the area of ​​the convex hull is calculated using the polygon area formula; The minimum engineering mining area threshold refers to the minimum area of ​​the convex hull constructed by all boreholes within the same ore blending unit.

6. The data-driven intelligent partitioning method for open-pit polymetallic mine ore blending units according to claim 1, characterized in that, In step 4, K-means clustering is performed based on the joint feature vector. The number of clusters is adaptively determined according to the total area of ​​the convex hull constructed from all boreholes in the blast pile, the minimum area threshold of the convex hull constructed from boreholes in a single ore mixing unit, and the effective clustering ratio. DBSCAN connected component detection is used to split, recode, and adjust the discrete point assignment of the spatially discrete clustering results.

7. The data-driven intelligent partitioning method for open-pit polymetallic mine ore blending units according to claim 6, characterized in that, The joint feature vector is weighted to construct a weighted combined feature vector, and then K-means clustering is performed. The weighting is to combine spatial coordinates and quality features according to weights.

8. The data-driven intelligent partitioning method for open-pit polymetallic mine ore blending units according to claim 6 or 7, characterized in that, The objective function of K-means clustering is expressed as follows: in, C k Indicates the first k A collection of boreholes for each ore blending unit. for C k The centroid in the combined feature space yes C k Inner i A feature vector that is a weighted combination of borehole spatial coordinates and quality features; Based on domain experience, the number of ore blending units is pre-defined as a decision variable. K The initial search range is The total area of ​​the convex hull of the computation region is A total Based on minimum area constraint A min The theoretical upper bound for estimating the number of clusters using the total area of ​​the convex hull of the regions is: K min This represents the minimum number of ore blending units. This represents the maximum number of ore blending units. The area utilization rate coefficient is the adjusted initial search interval. ; If clustering C k convex hull area ,but C k For effective clustering, let's call it . Otherwise, it is 0. Within, retain the cluster with the largest effective cluster ratio. K The value is used as the input parameter for K-means clustering; Calculate the spatial distance of borehole samples within the cluster, determine the DBSCAN neighborhood radius and minimum sample number, and then apply the DBSCAN algorithm to... C k Perform density clustering and output the connected component labels for each borehole sample. If the number of connected components is not less than 2, then it is considered... C k There exist spatial discrete subclusters; if it is 1, then C k Meets spatial connectivity requirements; If DBSCAN outputs multiple connected components, each connected component containing at least 3 boreholes is split into independent candidate units, and the candidate units are re-encoded. For discrete point sets with fewer than 3 boreholes, the distance to the centroid of each valid candidate unit is calculated, and the set is assigned to the nearest valid candidate unit.

9. The data-driven intelligent partitioning method for open-pit polymetallic mine ore blending units according to claim 1, characterized in that, The evolutionary algorithm uses fixed-length integer encoding to represent the affiliation relationship between boreholes and ore blending units, performs decoding constraint correction on the encoded individuals, calculates the fitness value including the quality uniformity objective function and constraint penalty terms, and performs selection, crossover and mutation operations.

10. The data-driven intelligent partitioning method for open-pit polymetallic mine ore blending units according to claim 1, characterized in that, In step 6, a Voronoi diagram is constructed using all borehole points as generators. The unbounded Voronoi units are finiteized and trimmed with the blast pile boundary polygons. The union of the borehole-level Voronoi units is then calculated according to the ore blending unit to which the borehole belongs, to obtain the ore blending unit-level polygon boundary.